Functional Group Identification for FTIR Spectra Using Image-Based Machine Learning Models.

Area of Science:

• Spectroscopy
• Machine Learning
• Computational Chemistry

Background:

• Fourier transform infrared spectroscopy (FTIR) is crucial for identifying functional groups in chemical compounds.
• Manual spectral interpretation is time-consuming and requires expertise.
• Developing automated methods can significantly improve analysis efficiency.

Purpose of the Study:

• To develop a generalizable machine learning model for automated functional group identification in gas-phase FTIR spectra.
• To reduce the time and effort required for FTIR spectral interpretation.
• To expand the applicability of FTIR measurements for organic sample analysis.

Main Methods:

• Utilized convolutional neural networks (CNNs), a type of machine learning algorithm.
• Acquired intensity-frequency data from 8728 gas-phase organic molecules from the NIST spectral database via web scraping.
• Transformed spectral data into images for input into the CNN models.

Main Results:

• Successfully trained CNN models capable of identifying 15 common organic functional groups.
• Validated model performance on previously unseen FTIR spectra.
• Demonstrated the ability of broad functional group models to infer in tandem for comprehensive spectrum interpretation.

Conclusions:

• The developed ML models significantly reduce analysis time for functional groups in FTIR spectra.
• This represents the first implementation of image-based CNNs for predicting functional groups from spectroscopic data.
• The approach facilitates facile analysis of organic samples using FTIR, enhancing its practical utility.